0.0.1 Dealing with duplicated ids

When evaluating if the dataset had duplicated ids, I found that id 257 was duplicated and there was no id 254. In the table below, I am just showing a few variables and participants from school 2.

After checking with CEQP’s research assistant, I corroborated that one of the duplicated cases of id 257 in fact was id 254. I fixed this mistake with the code below using the response_id variable and the mutate and case_when functions.

elt_w1_clean <- elt_w1_clean %>%
  mutate(id = case_when(response_id == "R_6EELe7Uuwi9W7zX" ~ "254",
                        TRUE ~ as.character(id))) %>%
  arrange(id)

0.0.2 Dealing with survey coding errors

The id protocol followed in CEQP projects is very straightforward. They usually use three digits for each individual participant id and use the first of these three digits to indicate the school id. In this system, ids in the 100’s would belong to school 1, ids in the 200’s to school 2, and so on.

By visual inspection I dentified that the first digit of the indvidual ids in the id variable did not correspond to the ids in the school id variable school for schools 3, 4, 5, and 6. In the table below, I selected four variables and only the first row of data of each of the six schools to ilustrate this point.

As can be seen in the table above, ids in the 300’s are coded to belong to school 4 and ids in the 400’s are coded to belong to school 3. I am calling this flip-flopped school ids. Schools 5 and 6 were also flip-flopped.

At first, I thought that this could be due to an error in the data exporting process and it seemed like an easy enough fix to make. I thought I just needed to recode the names of the levels of the school variable. Later I found that this fix did not solve the issue. It took me a couple of months to identify that the error was coded in the Qualtrics survey.

The images below are screenshots of the same raw data SPSS file downloaded directly from Qualtrics. In figure 1, it can be seen that when the value labels button is “on” (i.e. showing value labels and not values), it appears as if there was no flip-flop because the names of the schools coincided with the numbers that were assigned to them. Indeed, “Kelly” was school 3 and its participants were identified with ids in the 300’s and “ATA” was school 4 and its participants were identified with ids in the 400’s, and so on.

Figure 1: Value labels button on.

This changed when the value labels button was “off”. In the image below, the flip-flopped school ids is evident again:

Figure 2: Value labels button off

This survey coding error meant that the school variable’s value labels properly corresponded to the participants’ ids, but the variable’s values did not. Instead of recoding the values, I decided to create a new variable called school_id and delete the flawed original variable school.

In the code below, I used the first digit of the individual participant id variable id as the reference for the new school_id variable, following CEQP’S id protocol. I also created a new variable called condition to indicate which schools were randomly assigned to the control group (coded as 1) or to the intervention group (coded as 2).

I coded schools identified with a school_id odd number (1, 3, and 5) as the control schools and the schools identified with an even number (2, 4, and 6) as the intervention schools, as directed by CEQP’s research assistant. Finally, I also created a wave variable to indicate the wave of the data.

elt_w1_clean_2 <- elt_w1_clean %>%
  mutate(school_id = str_sub(id, 1, 1), # new school id variable
         condition = case_when(
           school_id == "1" | school_id == "3" | school_id == "5" ~ "1",
           school_id == "2" | school_id == "4" | school_id == "6" ~ "2")) %>% # new condition variable
  select(school_id, condition, everything()) %>%
  add_column(wave = 1, .before = 9) %>% # new wave variable
  select(- school) # deleting school variable

The condition and school_id variables I created in the previous code were string variables. In the code below I made them numeric so they can be used in quantitative analyses. I also added value labels with the set_vall function so that SPSS users can use the value labels button.

In the code below I also fixed a response option coding error I identified in the variable q68. Throughout most of the survey, response options were coded as “Strongly Disagree” = 1, “Disagree” = 2, “Agree” = 3, “Strongly Agree” = 4, “No response” = 99; however, in variable q68 the response option “No response” was coded as 5.

I fixed this using the ifelse function, specifying that if this variable had a response of 5, it should be changed to 99. Finally, I set the variable and value labels with the set_varl and set_vall functions, respectively, because sometimes procedures performed with R strips out these labels.

elt_w1_clean_3 <- elt_w1_clean_2 %>%
  mutate(condition = as.numeric(condition),
         condition = set_vall(condition, c("control" = 1, "intervention" = 2)),
         school_id = as.numeric(school_id),
         school_id = set_vall(school_id, c("cascade" = 1, "prairie_mountain" = 2, "kelly" = 3, "ata" = 4, "briggs" = 5, "agnes_stewart" = 6)),
         q68 = ifelse(q68 == 5, 99, q68),
         q68 = set_varl(q68, "When I communicate with Latino families, I keep in mind that many Latino parents may not understand how to navigate the educational system in this
country."),
         q68 = set_vall(q68, c("Strongly Disagree" = 1, "Disagree" = 2, "Agree" = 3, "Strongly Agree" = 4, "No response" = 99)))

0.0.3 Dealing with split out responses from multiple choice, unique answer variables

In this dataset, several multiple choice variables that were originally meant to have a single answer, were spread out as if they had multiple answers. I believe this was because in the Qualtrics survey development process, the option for Multiple answer was selected, instead of Single answer.

Figure 3: Qualtrics survey development

When this happens, participants could select mutually exclusive options, like this:

Figure 4: Qualtrics survey preview

When Multiple answer is selected, Qualtrics splits these multi-value fields into columns, assigning a value of 1 if a response option is chosen and a value of 0 if a response option is not chosen. In the following code, I collapsed the Spanish variable that was split out so it could be easily used in analyses. To avoid overwhelming the reader, I am omitting the code I used to collapse other language variables. I used the same procedure in all of these variables.

In the code below, the function pivot_longer makes the dataset “long” as it increases the number of rows and decreases the number of columns. This function gathers varables’ names under the new variable item_2 and gathers the values of these variables under the new variable spanish_comfort. Then, I chose only the options that had a value of 1, indicating when a participant chose that response option.

Finally, I recoded the response options to follow this scheme: “Not at all comfortable” = 1, “Somewhat comfortable” = 2, “Comfortable” = 3, “Very comfortable” = 4, “No response” = 99.

# collapsing spanish variables
spa <- elt_w1_clean_3 %>%
  select(id, starts_with("q132_2")) %>% # creating a dataset with only the id and Spanish variables
    pivot_longer(
      cols = starts_with("q132_2"), 
      names_to = "item_2",
      values_to = "spanish_comfort",
      values_drop_na = TRUE) %>% 
   filter(spanish_comfort == 1) %>%
  mutate(spanish_comfort = case_when(item_2 == "q132_2_1" ~ "1",
                                     item_2 == "q132_2_2" ~ "2",
                                     item_2 == "q132_2_3" ~ "3",
                                     item_2 == "q132_2_4" ~ "4",
                                     item_2 == "q132_2_99" ~ "99",
                        TRUE ~ as.character(spanish_comfort))) %>%
  select(-item_2) # selecting out variable with repetitive information

When all the language variables were collapsed I tested if there were duplicated cases and I found that participant identified with id 454 chose response option 1 and response option 2.

Because I can only assume that this was an entry error because the choices are mutually exclusive, “Not at all comfortable” = 1, vs. “Somewhat comfortable” = 2, I used the distinct function to retain only unique values.

For this case, option 1 = “Not at all comfortable” was retained as the function “assumes” the second option is the duplicative.

spa_2 <- spa %>% 
  distinct(id, .keep_all = TRUE)

The last step in this process was making the language variables numeric so they could be used in quantitative analyses and adding the value labels so that the SPSS users can use the value labels button. I used the code below to do this.

elt_w1_clean_4 <- elt_w1_clean_4 %>%
  mutate(english_comfort = as.numeric(english_comfort),
         english_comfort = set_vall(english_comfort, c("not at all comfortable" = 1, "somewhat comfortable" = 2, "comfortable" = 3, "very comfortable" = 4, "no response" = 99)),
         spanish_comfort = as.numeric(spanish_comfort),
         spanish_comfort = set_vall(spanish_comfort, c("not at all comfortable" = 1, "somewhat comfortable" = 2, "comfortable" = 3, "very comfortable" = 4, "no response" = 99)),
         other1_lang_comfort = as.numeric(other1_lang_comfort),
         other1_lang_comfort = set_vall(other1_lang_comfort, c("not at all comfortable" = 1, "somewhat comfortable" = 2, "comfortable" = 3, "very comfortable" = 4, "no response" = 99)),
         other2_lang_comfort = as.numeric(other2_lang_comfort),
         other2_lang_comfort = set_vall(other2_lang_comfort, c("not at all comfortable" = 1, "somewhat comfortable" = 2, "comfortable" = 3, "very comfortable" = 4, "no response" = 99)),
         )

As shown, the following variables were the result of the collapsing process described above: english_comfort, spanish_comfort, other1_lang_comfort, and other2_lang_comfort.

0.0.4 Renaming demographic variables

In the code below, I used the rename function to rename the demographic variables that I will use to describe participants’s characteristics in the next section of this report. This function uses a “new name” = “old name” pattern. Very straightforward!

At the end I selected out a few variables that did not have meaningful information. For instance, variable q127 was a response/no response question that only indicated if participants chose to answer it. The meaningul information was contained in variable q127_1_text that was renamed as age.

elt_w1_clean_5 <- elt_w1_clean_4 %>%
  rename(c("age" = "q127_1_text"), 
         c("birth_country" = "q128"),
         c("another_birth_country_text" = "q128_2_text"),
         c("age_first_moved_us" = "q129_1_text"),
         c("white" = "q130_1"),
         c("hispanic_latino_spanish" = "q130_2"),
         c("black_african_american" = "q130_3"),
         c("asian" = "q130_4"),
         c("american_indian_alaska_native" = "q130_5"),
         c("indigenous_americas" = "q130_6"),
         c("middle_eastern_north_african" = "q130_7"),
         c("native_hawaiian_pacific_islander" = "q130_8"),
         c("race_ethnicity_other" = "q130_9"),
         c("race_ethnicity_no_response" = "q130_99"),
         c("indigenous_americas_text" = "q130_6_text"),
         c("race_ethnicity_other_text" = "q130_9_text"),
         c("gender_id" = "q131"),
         c("years_in_position" = "q133"),
         c("years_in_school" = "q134"),
         c("equity_leadership" = "q135_1"),
         c("cultural_responsiveness" = "q135_2"),
         c("restorative_practices" = "q135_3"),
         c("diversity" = "q135_4"),
         c("ell" = "q135_5"),
         c("cont_ed_other" = "q135_6"),
         c("cont_ed_na" = "q135_88"),
         c("cont_ed_no_response" = "q135_99"),
         c("cont_ed_other_text" = "q135_6_text")) %>%
  mutate(age = as.numeric(age)) %>% # making it numeric for QUAN analyses
  select(-q127, -q129, -q131_3_text) # selecting out because they did not have meaningful info